Re: [PATCH] mm/memory: Fix missing pte marker for !page on pte zaps

[Peter Xu]

On Wed, Mar 13, 2024 at 11:03:04PM +0100, David Hildenbrand wrote:
> On 13.03.24 22:31, peterx@xxxxxxxxxx wrote:
> > From: Peter Xu <peterx@xxxxxxxxxx>
> > 
> > Commit 0cf18e839f64 of large folio zap work broke uffd-wp.  Now mm's uffd
> > unit test "wp-unpopulated" will trigger this WARN_ON_ONCE().
> 
> Good that I added the WARN_ON_ONCE() :)

To be explict, VM_WARN_ON_ONCE. :) And that's my guess that you didn't hit
it when you posted the series and did the tests, as I know latest distros
like Fedora dropped DEBUG_VM, so maybe you had your base config out of
there (but I normally have it irrelevant of that).

> 
> > 
> > The WARN_ON_ONCE() asserts that an VMA cannot be registered with
> > userfaultfd-wp if it contains a !normal page, but it's actually possible.
> > One example is an anonymous vma, register with uffd-wp, read anything will
> > install a zero page.  Then when zap on it, this should trigger.
> 
> Are you sure? zap_install_uffd_wp_if_needed() contains right at the start:
> 
> 	/* Zap on anonymous always means dropping everything */
> 	if (vma_is_anonymous(vma))
> 		return;

My example is not exactly how the test failed, but should be a simpler
version of it.  To trigger this warning I don't think it requires the zero
page to be wr-protected at all or have any pte marker involved.
UFFDIO_REGISTER should suffice, afaiu (feel free to read the example above
again; there's no mention of ioctl(UFFDIO_WRITEPROTECT)).

> 
> So if that's the case the unit test triggers, I'm confused.
> 
> > 
> > What's more, removing that WARN_ON_ONCE may not be enough either, because
> > we should also not rely on "whether it's a normal page" to decide whether
> > pte marker is needed.  For example, one can register wr-protect over some
> > DAX regions to track writes when UFFD_FEATURE_WP_ASYNC enabled, in which
> > case it can have page==NULL for a devmap but we may want to keep the marker
> > around.
> 
> I thought uffd-wp was limited to specific backends only. But looks like that
> changed with UFFD_FEATURE_WP_ASYNC, I guess?

Correct.  That was also what the new PAGEMAP ioctl relies on.

> 
> 
> Change itself looks, good. Not sure about the anon_vma example above.
> 
> Thanks!
> 
> Acked-by: David Hildenbrand <david@xxxxxxxxxx>

Thanks,

-- 
Peter Xu